BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Centralized resource scheduling and vertical integration in datace
 nter servers
DTSTART:20210628T090000
DTEND:20210628T110000
DTSTAMP:20260410T125202Z
UID:c81355ef96adf88b55c3a70732a9db60a2f5a215e1415e7887a50961
CATEGORIES:Conferences - Seminars
DESCRIPTION:Konstantinos Prasopoulos\nEDIC candidacy exam\nexam president:
  Prof. Katerina Argyraki\nthesis advisor: Prof. Edouard Bugnion\nco-examin
 er: Prof. Sanidhya Kashyap\n\nAbstract\nMicrosecond-scale datacenter appli
 cations spend a\nfew μs of processing per RPC and communcate using sub-\n
 10μs networking fabrics. However\, existing systems and abstractions\nadd
  disproportionate latency overheads. This document\npresents three systems
  that tackle the aforementioned issue.\nHoma and Shinjuku use centralised 
 schedulers at the host level\nto optimally schedule RPCs on networking and
  CPU resources\nrespectively. Snap addresses the issue through a microkern
 elinspired\nnetworking stack that presents a low-overhead interface\nto ap
 plications and efficiently manages CPU resources. Finally\,\nwe propose a 
 new transport protocol that handles congestion\ndifferently based on its o
 rigin.\n\nBackground papers\n1) Marty et al. "Snap: a microkernel approac
 h to host networking". https://dl.acm.org/doi/10.1145/3341301.3359657\n2) 
 Montazeri et al. "Homa: a receiver-driven low-latency transport protocol u
 sing network priorities". https://dl.acm.org/doi/10.1145/3230543.3230564\
 n3) Kaffes et al. "Shinjuku: preemptive scheduling for µsecond-scale tail
  latency". https://dl.acm.org/doi/10.5555/3323234.3323264\n\n 
LOCATION:
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR